Automatic classification of pore structures of low-permeability sandstones based on self-organizing-map neural network algorithm

Abstract

Objective The pore system of low-permeability sandstone reservoirs is intricate, and the distribution of pore-throat sizes is highly variable. The microscopic pore structure significantly influences the reservoir′s petrophysical properties and plays a critical role in controlling fluid flow within sandstone reservoirs. Traditional approaches for evaluating pore structures primarily rely on morphological analyses of pore throat size distributions or regression analyses of pore structure parameters. These methods are significantly affected by human bias and often lack precise evaluation frameworks. Methods Poroperm analysis, mercury injection capillary pressure, nuclear magnetic resonance (NMR) measurements, and X-ray computed tomography (X-ray CT) scanning experiments were performed to characterize the pore structures of the Es4s low-permeability sandstones in the G oilfield, Bohai Bay Basin. On this basis, 15 parameters that reflect the microscopic features of low-permeability sandstones were selected, and four types of pore structures were classified by applying an unsupervised self-organizing-map neural network algorithm. Results The findings reveal that the Type Ⅰ pore structure predominantly features large pore throats, with a median throat radius (r50) ranging from 0.38 to 2.35 μm. This type exhibits excellent pore connectivity, contributing significantly to permeability. The petrophysical properties and pore connectivity of Type Ⅱ pore structures are second only to those of Type Ⅰ pore structures. The movable fluid porosity ranges from 2.76% to 5.61%, and the median throat radius (r50) is primarily distributed in the range of 0.01 to 0.23 μm. Type Ⅲ pore structures display good pore connectivity along with considerable microscopic heterogeneity. The petrophysical properties and seepage properties of Type Ⅲ pore structures are comparable to those of Type Ⅰ and Type Ⅱ pore structures. The Type Ⅳ pore structures are characterized by small pore throats and poor microscopic connectivity, which hinders fluid movement within the sandstones. Conclusion The self-organizing map neural network algorithm effectively classifies pore structure types in cases involving multiple parameters. The classification results are not affected by inaccurate user-defined information, and there is no limitation on the number of parameters involved in the training process, making the application effect in pore structure classification remarkable. The established pore structure evaluation scheme, which is based on a self-organizing feature map neural network algorithm, is vital for investigating the microscopic seepage behavior and reservoir quality of low-permeability sandstones.

Keywords

• bohai bay basin
• low-permeability sandstone
• pore structure
• self-organizing map neural network
• unsupervised mode